PyTorch vs TensorFlow | Ishan Misra and Lex Fridman

Please support this podcast by checking out our sponsors:

GUEST BIO:
Ishan Misra is a research scientist at FAIR working on self-supervised visual learning.

PODCAST INFO:

SOCIAL:

[ogrendang]

Asking a Facebook employee about pytorch vs tensorflow. 🤔

[BapiKAR]

PyTorch is what I made a transition to from Tf2.x(x>=7). So far I see it quite cool & friendly to research community. And yes, I converted my entire TF code to PyTorch in less than 2weeks.

[Fordance]

TF usually has some features 1-2 year ahead pytorch. Pytorch is more flexible for tweaking the model. TF2.4-2.6 was very buggy with strange errors that took long time to fix. That was the time I switched to mainly pytorch. I think TF is better now.

[adityanjsg]

Tensorflow...!
Great support, model integration and model deployment.

[qedqubit]

wow 😀 ! you are diverse in your topics ! luv 'it 😘!

[ravipratapmishra]

For me pytorch can give you more granular control with small learning curve, but when it comes to deplotment and documentation tf is way ahead.

[osn]

I like calling pytorch imperative! Because it you would understand machine learning in better way! Unlike declarative approaches similar to Unix like command and SQL you give it a command or query and it would run. Even through that Tensorflow uses procedural language, coding with it similar to declaring command. And static graph structuring making it difficult especially for beginners debugging code!

[carlossegura]

PyTorch’s source code is elegant and well thought out.

[perceptron]

Both are great. TF with Keras has better performance, strong community support and robustness 👍PyTorch better for research and experimentation. Easier to use and debug since is more pythonic. Better Dynamic computation graph. Easier to deploy on web and mobile. Your choice will be determine on: use case and developer's preferences.

[Ikaikaalika]

pytorch is more intuitive for SWEs, tensorflow is killer now with the keras integration

[GChief]

PyTorch has more applications over tensorflow

[stephennfernandes]

Pytorch Forever 🔥🔥🔥 ... I know the XLA TPU support isn't almost there yet but yeah Pytorch 🔥

[totheknee]

More like collaboration than competition. The open source _COMMUN_-ity shows how a _commune_ environment that _shares_ resources can be so vastly superior to the capitalistic wealth redistribution scheme, it isn't even funny.

[BobF]

This content is incredibly moving. I read a book with akin material that reshaped my worldview. "AWS Unleashed: Mastering Amazon Web Services for Software Engineers" by Harrison Quill

[arods]

Very respectfully, but it's quite hard to understand what he is telling.

[tashinahmed]

doesn't answer the question at all :|

[MrGeorgetudo]

WHY THE FUCK IS LEX FRIDMAN TALKING ABOUT MACHINE LEARNING

[gorgolyt]

Pretty useless clip tbh, all he says is "I prefer Pytorch because I've been using it for longer". He mentions that the imperative style is easier to debug, but TensorFlow 2 also uses an imperative style.

[davisonyeoguzoro]

Bias
Try asking a Google employee the same question